Nvidia Offers Guarantees to Cloud Providers in Exchange for Revenue Sharing
N.R. Finch
Nvidia is guaranteeing to buy back unsold GPU capacity from smaller cloud providers in exchange for a share of their cloud revenue, with disclosed commitments already reaching billions of dollars. This means → Nvidia is shifting from selling chips to underwriting downstream income, using its own balance sheet to reshape how AI compute gets financed.
How does this deal actually work?
Nvidia internally calls it the "AI Compute Partnership." The core term: if a cloud provider can't rent out its GPUs, Nvidia buys back the idle capacity at a fixed price.
In return, the cloud provider gives Nvidia a cut of its cloud revenue. That cut decreases over the life of the contract.
In plain terms = Nvidia acts as a safety net — it absorbs the downside if demand falls short, but takes a slice of the upside when business is good.
Why would Nvidia backstop someone else's risk?
GPUs are the most expensive component in an AI data center. Smaller providers have low credit ratings and struggle to secure loans.
With Nvidia's guarantee, lenders see a backstop in the worst case. One data center executive called it a "two birds, one stone" move — solving GPU financing and facility financing at once.
This means → Nvidia is lending its credit to help buyers unlock leverage they couldn't access alone. In effect, it is turning chip sales into a financial product.
How much money is already committed?
Last September, Nvidia pledged $630 million to buy back all of CoreWeave's unsold capacity through 2032. CoreWeave's stock rose nearly 30% the following week.
By the quarter ending April this year, Nvidia disclosed an additional $3.5 billion in guarantees backing customer data-center leases, in exchange for the right to subscribe for their shares.
Nvidia has also made equity investments in emerging cloud providers including CoreWeave and Lambda, and leased GPU servers back from them — totaling billions of dollars.
What is the strategic anxiety behind this?
Amazon, Microsoft, Meta, Google, and a handful of other hyperscalers currently buy the vast majority of Nvidia's chips — and these same companies are developing competing AI chips in-house.
This reflects Nvidia's core concern: its biggest customers are becoming its biggest rivals. Supporting mid-tier providers like CoreWeave is fundamentally about diversifying customer-concentration risk.
In plain terms = Nvidia doesn't want all its eggs in a few hyperscalers' baskets — especially when those hyperscalers are building their own baskets.
How far can this model scale?
Nvidia is currently in talks to guarantee leases for a major data-center project in Ohio. If fully built at current prices for chips, labor, and power, the project would cost $500 billion.
Whether Nvidia is willing to extend its guarantee model to a project of that scale remains the key variable the market is watching.
This means → if Nvidia does backstop mega-scale projects, its role shifts entirely from chip supplier to financial guarantor of AI infrastructure — with both risk and reward jumping by orders of magnitude.
Content is for reference only, not financial advice.